Investigating Writer-Independent Deep Learning Techniques for Offline Handwritten Signature Verification

Research proposes a deep learning-based approach for writer-independent authentication of handwritten signatures. Though handwritten signatures prove to be one of the most popular methods for biometric authentication, verification of the same is quite challenging because of the high variability in the writing styles across subjects. This paper proposes a new way of extracting relevant features from signature images and learning writer-independent representations of signatures through Convolutional Neural Networks. We evaluate our approach on the publicly available dataset of CIDER and demonstrate that such an approach outperforms the state-of-the-art techniques in both aspects: accuracy and robustness against different forgeries. Our results imply a rather promising approach towards practical applications, as a CNN-based system will lead to a very satisfying solution for signature verification.